Table 4.

Accuracy of models with original fire images (6000 images).

Models	Training Input Size	Training (AP50)	Testing Input Size	Testing (AP50)	Training Time	Iteration Number
YOLOv4 [17]	416 × 416	68.5%	608 × 608	65.7%	78 h	150
YOLOv4-tiny [17]		46.2%		43.4%	9 h
YOLOv3 [48]		61.8%		58.6%	86 h
YOLOv3-tiny [48]		40.2%		36.4%	12 h
Improved YOLOv4		69.3%		67.9%	74 h